Method and system for localization of faults in an industrial manufacturing plant

Abstract

Existing systems for fault detection and classification have the disadvantage that they have limited or no capability for fault localization and root cause identification, probably due to the challenges associated with modeling the nonlinear interactions among process variables and capturing the nonstationary behavior that is typical of most industrial processes. The disclosure herein generally relates to industrial manufacturing systems, and, more particularly, to method and system for localization of faults in an industrial manufacturing plant. The system uses a perturbation based approach for fault localization, in which the system determines variables having dominant effect on identified faults, in terms of a perturbation score calculated for each of the variables.

What is claimed is: 
     
         1 . A processor implemented method, comprising:
 receiving, via one or more hardware processors, a plurality of multivariate time-series data corresponding to operations of an industrial manufacturing plant, as input data, wherein the plurality of multivariate time-series data comprise information from a plurality of sensors monitoring working of a plurality of components of the industrial manufacturing plant;   pre-processing, via the one or more hardware processors, the input data to generate a pre-processed data;   computing, via the one or more hardware processors, a plurality of soft-sensor data using the pre-processed data, wherein a plurality of soft-sensors corresponding to the computed soft-sensor data includes physics-based soft sensors and data-driven soft sensors;   combining, via the one or more hardware processors, the pre-processed data with the plurality of soft-sensor data to generate a combined plant data;   performing, via the one or more hardware processors, a fault analysis on the combined plant data, comprising a) determining whether one or more faults are present in the combined plant data, and b) identifying a fault detection point if the one or more faults are present in the combined plant data; and   performing, via the one or more hardware processors, a fault localization to determine one or more variables from the combined plant data having a dominant effect on the one or more faults, if the combined plant data is determined to have the one or more faults, wherein performing, the fault localization comprises applying a multi-level variable perturbation, comprising:
 perturbing one or more of the variables from the combined plant data, input to a selected diagnosis model across the fault detection point to calculate a perturbation score corresponding to each of the one or more of the variables, in a plurality of iterations; 
 computing a contribution score as equal to value of a cumulative perturbation score of the plurality of iterations, and sorting the plurality of variables in the order of the contribution score; 
 eliminating all variables from among the one or more variables, having value of the contribution score below a predetermined percentile, to obtain a plurality of candidate variables, wherein the plurality of candidate variables are variables having the dominant effect on the one or more faults; and 
 iteratively performing perturbation on the plurality of candidate variables, a) for a predetermined number of perturbation levels, or b) till a predetermined number of localized variables are obtained. 
   
     
     
         2 . The method of  claim 1 , wherein the perturbation score for each of the one or more variables is calculated in each of the plurality of iterations, wherein in each of the plurality of iterations, the perturbation score is obtained as output of the diagnostic model after replacing a faulty data with corresponding normal data for the one or more variables. 
     
     
         3 . The method of  claim 1 , wherein performing the fault analysis comprises classifying one or more detected faults into one or more known fault classes by a fault classification model, wherein, the fault classification model is retrained if failed to classify any detected fault into the one or more known fault classes. 
     
     
         4 . The method of  claim 1 , wherein the diagnosis model is selected from among a plurality of pre-trained diagnosis models for the one or more fault classes the one or more detected faults are classified into. 
     
     
         5 . The method of  claim 1 , wherein the plurality of multivariate time-series data comprise a plurality of sensor data, laboratory information, and environment data. 
     
     
         6 . The method of  claim 1 , wherein the number of variables perturbed simultaneously at each of the plurality of iterations is equal to the corresponding perturbation level. 
     
     
         7 . The method of  claim 1 , wherein the perturbation level is a positive integer not exceeding number of candidate variables from a preceding level from among a plurality of preceding levels. 
     
     
         8 . The method of  claim 1 , wherein the fault detection point is a time instance of occurrence of the fault. 
     
     
         9 . A system, comprising:
 one or more hardware processors;   a communication interface; and   a memory storing a plurality of instructions, wherein the plurality of instructions when executed, cause the one or more hardware processors to:
 receive a plurality of multivariate time-series data corresponding to operations of an industrial manufacturing plant, as input data, wherein the plurality of multivariate time-series comprise information from a plurality of sensors monitor working of a plurality of components of the industrial manufacturing plant; 
 pre-process the input data to generate a pre-processed data; 
 compute a plurality of soft-sensor data using the pre-processed data, wherein a plurality of soft-sensors corresponding to the computed soft-sensor data includes physics-based soft sensors and data-driven soft sensors; 
 combine the pre-processed data with the plurality of soft-sensor data to generate a combined plant data; 
 perform a fault analysis on the combined plant data, by a) determining whether one or more faults are present in the combined plant data, and b) identifying a fault detection point if the one or more faults are present in the combined plant data; and 
 perform a fault localization to determine one or more variables from the combined plant data having a dominant effect on the one or more faults, if the combined plant data is determined to have the one or more faults, wherein performing, the fault localization comprises applying a multi-level variable perturbation, by:
 perturbing one or more of the variables from the combined plant data, input to a selected diagnosis model across the fault detection point to calculate a perturbation score corresponding to each of the one or more of the variables, in a plurality of iterations; 
 computing a contribution score as equal to value of a cumulative perturbation score of the plurality of iterations, and sorting the plurality of variables in the order of the contribution score; 
 eliminating all variables from among the one or more variables, having value of the contribution score below a predetermined percentile, to obtain a plurality of candidate variables, wherein the plurality of candidate variables are variables having the dominant effect on the one or more faults; and 
 iteratively performing perturbation on the plurality of candidate variables, a) for a predetermined number of perturbation levels, or b) till a predetermined number of localized variables are obtained. 
 
   
     
     
         10 . The system of  claim 9 , wherein the one or more hardware processors are configured to calculate the perturbation score for each of the one or more variables in each of the plurality of iterations, wherein in each of the plurality of iterations, the perturbation score is obtained as output of the diagnostic model after replacing a faulty data with corresponding normal data for the one or more variables. 
     
     
         11 . The system of  claim 9 , wherein the one or more hardware processors are configured to perform the fault analysis by classifying one or more detected faults into one or more known fault classes by a fault classification model, wherein, the fault classification model is retrained if failed to classify any detected fault into the one or more known fault classes. 
     
     
         12 . The system of  claim 9 , wherein the one or more hardware processors are configured to select the diagnosis model from among a plurality of pre-trained diagnosis models for the one or more fault classes the one or more detected faults are classified into. 
     
     
         13 . The system of  claim 9 , wherein the plurality of multivariate time-series data comprise a plurality of sensor data, laboratory information, and environment data. 
     
     
         14 . The system of  claim 9 , wherein the number of variables perturbed simultaneously at each of the plurality of iterations is equal to the corresponding perturbation level. 
     
     
         15 . The system of  claim 9 , wherein the perturbation level is a positive integer not exceeding number of candidate variables from a preceding level from among a plurality of preceding levels. 
     
     
         16 . The system of  claim 9 , wherein the fault detection point is a time instance of occurrence of the fault. 
     
     
         17 . One or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause:
 receiving a plurality of multivariate time-series data corresponding to operations of an industrial manufacturing plant, as input data, wherein the plurality of multivariate time-series data comprise information from a plurality of sensors monitoring working of a plurality of components of the industrial manufacturing plant;   pre-processing the input data to generate a pre-processed data;   computing a plurality of soft-sensor data using the pre-processed data, wherein a plurality of soft-sensors corresponding to the computed soft-sensor data includes physics-based soft sensors and data-driven soft sensors;   combining the pre-processed data with the plurality of soft-sensor data to generate a combined plant data;   performing a fault analysis on the combined plant data, comprising a) determining whether one or more faults are present in the combined plant data, and b) identifying a fault detection point if the one or more faults are present in the combined plant data; and   performing a fault localization to determine one or more variables from the combined plant data having a dominant effect on the one or more faults, if the combined plant data is determined to have the one or more faults, wherein performing, the fault localization comprises applying a multi-level variable perturbation, comprising:
 perturbing one or more of the variables from the combined plant data, input to a selected diagnosis model across the fault detection point to calculate a perturbation score corresponding to each of the one or more of the variables, in a plurality of iterations; 
 computing a contribution score as equal to value of a cumulative perturbation score of the plurality of iterations, and sorting the plurality of variables in the order of the contribution score; 
 eliminating all variables from among the one or more variables, having value of the contribution score below a predetermined percentile, to obtain a plurality of candidate variables, wherein the plurality of candidate variables are variables having the dominant effect on the one or more faults; and 
 iteratively performing perturbation on the plurality of candidate variables, a) for a predetermined number of perturbation levels, or b) till a predetermined number of localized variables are obtained. 
   
     
     
         18 . The one or more non-transitory machine-readable information storage mediums of  claim 17 , wherein the perturbation score for each of the one or more variables is calculated in each of the plurality of iterations, wherein in each of the plurality of iterations, the perturbation score is obtained as output of the diagnostic model after replacing a faulty data with corresponding normal data for the one or more variables. 
     
     
         19 . The one or more non-transitory machine-readable information storage mediums of  claim 17 , wherein performing the fault analysis comprises classifying one or more detected faults into one or more known fault classes by a fault classification model, wherein, the fault classification model is retrained if failed to classify any detected fault into the one or more known fault classes. 
     
     
         20 . The one or more non-transitory machine-readable information storage mediums of  claim 17 , wherein the diagnosis model is selected from among a plurality of pre-trained diagnosis models for the one or more fault classes the one or more detected faults are classified into.